Nowadays, Wireless Sensor Networks (WSN) are widely been employed to solve agricultural problems related to the optimization of scarce farming resources, decision making support, and land monitoring. However, the small sensing devices that are part of WSNs – known as sensor nodes – suffer from degradation and so producing erroneous measurements. In this paper, a machine learning method based on Non-Negative Matrix Factorization (NMF) is applied to the spectral representation of data acquired by a WSN to extract features that model the normal behavior of sensor node readings leading to a good representation of data using a low number of features. This procedure is accompanied by a classifier that decides if there is a set of features that deviates from the normal ones. Experiments on soil moisture data show that NMF achieves good results detecting flaws in readings from sensors. Results are compared with other method based on Principal Component Analysis (PCA), the Multi-scale PCA (MSPCA) algorithm.
Bibliographical noteFunding Information:
This work has been partially supported by the Peruvian Government grant PITEI-1-P-275-092-14 (Inn?vate Per?) and the Universidad Cat?lica San Pablo.
This work has been partially supported by the Peruvian Government grant PITEI-1-P-275-092-14 (Innóvate Perú) and the Universidad Católica San Pablo .
© 2018 Elsevier B.V.
- Discrete Wavelet Transform (DWT)
- Non-Negative Matrix Factorization (NMF)
- Principal Components Analysis (PCA)
- Sensor nodes fault detection
- Wireless Sensor Networks (WSN)